ChatGPT for Business Pricing: What It Really Costs a Team
Finance just forwarded a renewal notice. Three people expensed ChatGPT on personal cards last quarter, sales wants it for drafting outreach, and support has been quietly pasting tickets into a chat window for months. So someone says the obvious thing: let's just get the team on a real plan. You open OpenAI's pricing page to pick a tier, and what looked like a $20-versus-$25 decision turns into a longer question. Which plan, for how many people, and does buying seats even solve the thing that made you look in the first place?
The seat price is the easy part, and it is cheap. The expensive part is everything the seat price does not cover: the work that repeats hundreds of times a day, the output nobody checks, the moment a chat window should have been a workflow. Buy seats to solve that and you have bought a nicer window for the person still doing the job by hand. The real cost lives one layer down, and it is worth seeing before you sign the renewal.
The ChatGPT business plans, and what each costs in 2026
OpenAI sells ChatGPT to companies in two shapes: a per-seat subscription for people who chat with it, and a per-token API for software that calls it. Start with the seats, because that is what most teams price first.
There are three tiers that matter for a business. Per OpenAI's business pricing page:
| Plan | Price (billed monthly) | Price (billed annually) | Seat minimum | |---|---|---|---| | Plus | $20 / user / month | not offered annually | 1 | | Business | $25 / user / month | $20 / user / month | 2 | | Enterprise | quote only | quote only | contact sales |
A few things are worth pinning down, because the naming has shifted. ChatGPT Business is the current name for the shared-workspace plan OpenAI used to call Team. It is the one most 1-to-50-person companies land on: a shared admin console, member management, and by default your workspace data is not used to train OpenAI's models, which is the line that matters if anyone is pasting in customer records. OpenAI also cut standard seat prices by $5 a month in April 2026, which is why Business now starts at $20 annually rather than the $25 it launched at, per OpenAI's pricing help article.
Plus is a single-user plan. It has no admin controls and no shared workspace, so using it as a "team" plan means everyone runs their own island with their own billing and no visibility for you. Fine for one founder. A mess at eight people.
Enterprise has no public per-seat price. OpenAI directs you to sales for a quote, and the widely circulated per-seat and seat-minimum numbers for it come from resellers and roundup blogs, not OpenAI, so treat them as rumor. For most lean teams Enterprise is not the question anyway. It is built for large deployments with procurement and legal in the room.
What the seat price actually buys, and where it stops
For ad-hoc knowledge work, seats are a good deal. A support lead who drafts replies, a marketer who rewrites landing copy, an ops person who cleans a messy export: each of them gets real value from a chat window, and $20 to $25 a month per person is nothing against the hours it saves. If your whole use of AI is people typing into a box and reading the answer, buy Business, turn on the admin controls, and you are done. We wrote the setup end to end in the ChatGPT for business setup guide.
The seat model breaks the moment the work stops being ad-hoc. A seat prices one human doing one thing at a time. It does not price a job that runs 400 times a day whether or not a person is watching. When support is pasting the same ticket-triage prompt hundreds of times, you are not short on seats. You are paying a person to be the copy-paste bridge between the inbox and the model, and no number of extra seats fixes that. The ChatGPT automation guide walks the exact point where a hand-run prompt should become an unattended workflow.
That is the real fork in ChatGPT pricing. Not $20 versus $25. It is seats versus API.
The API: per-token pricing, and why it is the cheap part
Anything automated runs on the API, not on seats. Instead of paying per person per month, you pay per token, roughly per word, split between the text you send in and the text the model sends back. Current published rates from OpenAI:
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Good for | |---|---|---|---| | GPT-5 | $1.25 | $10.00 | hard reasoning, messy judgment calls | | GPT-4.1 | $2.00 | $8.00 | reliable structured extraction and drafting | | GPT-4o mini | $0.15 | $0.60 | high-volume classification, tagging, routing |
The headline is how small these numbers are. A million tokens is roughly 750,000 words. GPT-4o mini reads 750,000 words of your input for 15 cents. Even GPT-5, the expensive one, charges $1.25 to read that same wall of text. For the vast majority of business tasks, the model itself is not where your money goes.
Which flips the usual worry on its head. Teams new to this ask how much the AI will cost. On per-task work the AI is a rounding error. The cost lives somewhere else entirely, and a worked example makes it obvious.
Worked example: triaging 40 support emails an hour
Take a concrete job. Every inbound support email gets read, tagged by intent, and given a draft reply in your tone. Say 40 an hour across the day, about 7,000 a month. One email plus the instructions runs maybe 800 input tokens, and a drafted reply about 300 output tokens. Do the math against the rates above.
- On GPT-4o mini: 800 input tokens costs $0.00012, 300 output tokens costs $0.00018. That is $0.0003 an email, or about $2 a month for all 7,000.
- On GPT-5: the same email runs about $0.004, or roughly $28 a month for the full volume.
So the model bill for automating a genuinely painful, all-day support task lands somewhere between $2 and $30. That is not a typo. Now weigh that against the alternatives. Buying Business seats does not automate this at all: it just gives the human doing the pasting a nicer chat window, and the person is still the bottleneck. Hiring someone to sit on the queue is a real salary. The token cost is the cheapest line on the page by a wide margin.
The expensive part is the part the token price hides. Something has to receive each email, call the model, force the output into clean structured data your helpdesk can read, catch the runs where the model is wrong before a bad reply goes out, retry on failures, and alert someone when it quietly breaks at 3am. That build, and keeping it alive, is the real cost of "ChatGPT for business" once you cross from chatting to automating. Not the $20 seat. Not the $2 token bill. The reliable workflow wrapped around it.
Where working with bottta changes the math
This is the call we make with clients constantly, so here is the straight version. If your use of ChatGPT is people typing into a box, do not overthink it. Buy ChatGPT Business, switch on the admin and data controls, and spend your energy on prompts and shared instructions, not on us. That is a solved problem and it is not worth a project.
The moment a task repeats at volume with no human reading each result, seats and DIY glue stop being the answer, and that is the work bottta is built for. We design and build the automation around the model: the trigger that fires on each new record, the API call to the right model at the right price tier, the structured output your CRM or helpdesk can actually ingest, the confidence gate that routes shaky answers to a human instead of shipping them, and the monitoring that tells you when it fails instead of you finding out from a customer. The token cost stays tiny. What you are paying for is a workflow that runs unattended and does not quietly rot.
Two ways to work with us. A $4K fixed-scope project when you have one clear automation to stand up, integrations included, with 30 days of post-launch support to shake out the edge cases. A $3K/month retainer when you have a stack of these and want a team owning up to three active workflows at a time, monitoring them and fixing them as your tools and volume change. Both are AI Automation and Integrations work: LLM routing and extraction wired into the systems you already run. If you are weighing this against doing it in-house or buying an off-the-shelf tool, the build vs buy vs hire breakdown runs the numbers on all three, and the AI automation guide for small teams covers what to hand off first.
A quick honesty note on the alternatives. You can wire a simple version of this yourself with Zapier or Make and an OpenAI key, and for a low-stakes, low-volume task you probably should. The DIY route falls down on exactly the parts that matter at scale: error handling, output validation, and monitoring nobody built because the happy path worked in the demo. That is the difference between a prompt that runs and a workflow you can trust with customer-facing output.
Frequently asked questions
How much is ChatGPT for a business per user?
ChatGPT Business is $25 per user per month billed monthly, or $20 per user per month billed annually, with a two-seat minimum, per OpenAI's business pricing page. Plus is $20 per month for a single user with no admin tools. Enterprise is quote-only through OpenAI's sales team.
Is ChatGPT Business the same as the old Team plan?
Yes, in practice. ChatGPT Business is the current name for the shared-workspace seat plan OpenAI previously marketed as Team. It carries the admin console, member management, and the default that your workspace data is not used to train OpenAI's models.
Do I pay per seat or per token?
Both exist, for different things. People who chat with ChatGPT are covered by per-seat plans like Business. Software that calls ChatGPT automatically is billed per token through the API, at rates like $0.15 per million input tokens for GPT-4o mini or $1.25 for GPT-5, per OpenAI's published pricing. Automating a recurring task uses the API, not extra seats.
Why is my automated ChatGPT bill so low but the project still costs money?
Because the model is the cheap part. Token costs for most per-task business jobs run from a couple of dollars to a few tens of dollars a month. The cost is in building and maintaining the reliable workflow around the model: triggers, structured output, error handling, a human-review gate, and monitoring. That engineering is what you are actually paying for, whether you build it in-house or bring in a studio.
Should a small team buy Enterprise?
Almost never. Enterprise is built for large deployments with procurement and legal involved, and it has no public price. A 1-to-50-person company is served by ChatGPT Business for chat and the API for anything automated. If your reason for wanting Enterprise is reliability or compliance on a specific workflow, the answer is usually a properly built automation, not a bigger seat plan.
Here is the test worth applying before the renewal. Count how many times a day someone runs the same prompt by hand. If the answer is once or twice, buy the seat and move on. If it is hundreds, the seat was never the fix, and the money should go toward the workflow that retires the copy-paste for good. Book a call with bottta if that second number is the one describing your team.